Stage-Dependent, Modality-Specific Coherence Regimes and Transition-Aligned Dynamics in Human Sleep

Stage-Dependent, Modality-Specific Coherence Regimes and Transition-Aligned Dynamics in Human Sleep Ankur Bhasin Bhasin Research Unit for Hyperphysics (BRUH) 26 December, 2025 Abstract Neural dynamics during sleep are commonly summarized by stage classification and bandlimited power, yet multichannel coherence patterns may reorganize in ways not captured by power alone. Here we map sleep EEG into a discrete coherence-state space and model the resulting sequences as session-level Markov chains. Using paired polysomnography (PSG) and cEEGrid recordings from OpenNeuro ds005207, we identify two reproducible session regimes (Normal and Coherence-shifted) after correcting pathological behavior of time-shift surrogate nulls via surrogate-variance regularization and confirming robustness using Markov-based surrogate families. Coherence-shifted sessions show modest but structured stage dependence, modality dissociation (PSG versus cEEGrid), and transition-aligned dynamical signatures, especially in cEEGrid. These results suggest that sleep coherence is not static within stage labels, but exhibits regime structure that is stage-conditional, modality-dependent, and coupled to stage transitions. 1 Introduction Sleep architecture is typically described as a sequence of discrete physiological stages (Wake, N1, N2, N3, REM). While stage labels capture canonical spectral and behavioral changes, they compress multichannel dynamics into a single categorical variable. Coherence and synchronization across channels are candidate descriptors of large-scale organization that may reorganize across stages and transitions, and may differ by measurement geometry (e.g., PSG versus cEEGrid). We test four concrete hypotheses: whether coherence dynamics during sleep (i) form reproducible session-level regimes, (ii) show stage-conditional structure, (iii) differ across recording modalities, and (iv) exhibit dynamical signatures aligned to stage transitions. 2 Methods 2.1 Dataset We analyzed overnight recordings from OpenNeuro ds005207 containing paired PSG and cEEGrid EEG. Sleep stages were extracted from the provided scoring event files and mapped to contiguous stage intervals in seconds. All analyses were performed per session and per modality. 2.2 Coherence-state construction EEG was segmented into fixed-length windows. For each window, coherence features were computed and summarized into a low-dimensional coherence representation, then discretized into six coherence 1 states (S0–S5). State S0 represents low-quality or noisy windows; states S1–S5 represent increasing coherence and cross-channel synchronization. 2.3 Markov modeling and entropy For each session and modality, the discrete state sequence was modeled as a first-order Markov chain. The Markov entropy rate Hwas computed for each session, and compared to surrogate distributions to assess deviation from null expectations. 2.4 Surrogate regularization and alternative nulls Time-shift surrogates can yield near-zero surrogate variance in long quasi-stationary sequences, producing inflated Z-scores. To prevent pathological inflation, surrogate variance flooring was applied: Zreg =Hdata −µsur max(σsur, σmin). We additionally generated Markov-based surrogate families (transition-matrix shuffles and degreepreserving Markov nulls) to test robustness under null models that preserve transition structure more directly than time shifts. 2.5 Regime classification Sessions were classified as Normal or Coherence-shifted based on regularized entropy deviations relative to surrogates, using conservative thresholds. 2.6 Stage-conditional and transition-aligned analyses We computed: •P(Coherence-shifted |stage,modality) with bootstrap confidence intervals •Transition-aligned instability change ∆instability = post −pre •Transition-aligned state occupancy deltas ∆occ(Si) All statistics were computed per session and aggregated at the population level, with bootstrap confidence intervals and FDR correction where applicable. 3 Results 3.1 Session-level example Figure 1 shows an example session summary, including the hypnogram with regime overlay, stageconditional regime probabilities, and transition-aligned dynamics. 3.2 Population-level stage dependence and modality dissociation Figure 2A shows population stage density across normalized sleep time with expected regime overlays. Stage-conditional probabilities (Figure 2B) show structured and modality-specific differences: PSG shows modest enrichment of Coherence-shifted regimes in N3 and suppression in REM, whereas cEEGrid shows depletion in N2 with relatively higher values in N1 and REM. 2 Table 1: Stage-conditional probability of Coherence-shifted regime by modality. Values are sessionlevel means with bootstrap 95% confidence intervals. Odds ratios (OR) are computed relative to the modality-specific baseline. Modality Stage P(shifted) 95% CI OR vs baseline PSG N1 0.28 [0.20, 0.35] 0.89 PSG N2 0.29 [0.25, 0.33] 0.94 PSG N3 0.35 [0.30, 0.41] 1.26 PSG REM 0.25 [0.21, 0.30] 0.78 cEEGrid N1 0.37 [0.29, 0.45] 1.14 cEEGrid N2 0.27 [0.22, 0.32] 0.71 cEEGrid N3 0.33 [0.29, 0.37] 0.98 cEEGrid REM 0.35 [0.27, 0.44] 1.07 3.3 Transition-aligned dynamics Figure 2C shows transition-aligned instability changes. PSG effects are near zero across transitions, whereas cEEGrid shows larger, regime-dependent ∆instability, suggesting stronger transitioncoupled dynamics in cEEGrid. 3.4 State occupancy reconfiguration State-occupancy deltas showed small effects overall. The strongest cEEGrid Coherence-shifted trend was a post-transition depletion of the hyper-coherent S5 state. This effect is consistent across bootstrap estimates but should be treated conservatively where it does not survive FDR correction. Table 2: Transition-aligned state occupancy changes (∆ occupancy = post - pre). Values shown are mean bootstrap estimates with 95% confidence intervals. Only states with consistent directional effects are shown. Modality Regime State Mean ∆ 95% CI cEEGrid Coherence-shifted S3 +0.0076 [-0.0005, 0.0156] cEEGrid Coherence-shifted S4 +0.0065 [-0.0009, 0.0137] cEEGrid Coherence-shifted S5 -0.0115 [-0.0225, -0.0014] PSG Coherence-shifted S3 -0.0047 [-0.0082, -0.0014] 4 Discussion These results suggest that coherence during sleep exhibits reproducible session-level regimes that are stage-conditional and modality-dependent. The strongest dynamical signatures occur around stage transitions and are more pronounced in cEEGrid than PSG, consistent with a measurementgeometry dependence in accessible coherence structure. Importantly, these effects persist after correcting known failures of time-shift surrogate nulls and after validating against Markov-based surrogate families, arguing against trivial temporal-autocorrelation artifacts. 3 5 Limitations This study is constrained by the modest number of sessions in ds005207 and by reliance on a specific discretization of coherence into six states. The mapping from continuous coherence features to discrete states may influence transition counts and entropy estimates, and different windowing parameters may shift effect sizes. Although Markov-based surrogates provide stronger nulls than time shifts, causal interpretations are not warranted here. Finally, some state-occupancy effects are small and may not survive correction in this dataset; replication in larger cohorts will be required to establish these patterns robustly. In addition, coherence states are derived from functional measures and do not imply direct anatomical or causal interactions. 6 Conclusion Sleep coherence is not static within stage labels. It exhibits structured regime dynamics that depend on stage, modality, and transitions. This framework provides a quantitative starting point for characterizing coherence organization beyond conventional staging, with potential utility for mechanistic and comparative sleep research. Figure 1: Example session-level analysis showing hypnogram with regime overlay, stage-conditional regime probabilities, and transition-aligned dynamics. 4 Figure 2: Population-level stage dependence, modality dissociation, and transition-aligned dynamics of coherence regimes. 5 References [1] Iber C, Ancoli-Israel S, Chesson A, Quan SF. The AASM Manual for the Scoring of Sleep and Associated Events. American Academy of Sleep Medicine, 2007. [2] Bullmore E, Sporns O. Complex brain networks: graph theoretical analysis of structural and functional systems. Nat Rev Neurosci. 2009;10:186–198. [3] Stam CJ. Modern network science of neurological disorders. Nat Rev Neurosci. 2014;15:683–695. [4] Hlinka J, Coombes S. Using computational models to relate structural and functional brain connectivity. Eur J Neurosci. 2012;36:2137–2145. [5] Deco G, Jirsa VK. Ongoing cortical activity at rest: criticality, multistability, and ghost attractors. J Neurosci. 2012;32:3366–3375. 6